[FFmpeg-devel] Possible cache use improvements

[Uoti Urpala]

On Sun, 2008-08-10 at 21:37 +0200, Michael Niedermayer wrote:
> Its very nice to see someone pick up arpis old idea of rowwise(=slice for
> mpeg2) decoding.
> Once you make it actually work and if its still faster then, then iam very
> interrested in this. Currently though its just 5% faster way to produce
> corrupted images this is not that usefull ...
> 
> Also i do not think a scratch area will be the only thing that has to be
> added t make this work. You likely will have to do something with the
> mb_skip code as well.

I think the mb_skip code would have to be disabled; it seems hard to
combine it in a way that would improve performance. My test video had
few still blocks so that wouldn't make much difference, and using a
proper scratch area vs the first row is unlikely to affect performance.
So a version without any image corruption will likely be equally fast in
that test case. In a video where a large portion of the image stays
unchanged the skip code could give more benefit; I don't know how
completely still the image would have to be for it to be an overall win.
However optimizing the already fast case like the skip logic does would
mostly be useful if you implement a decoded video buffer in the player
that allows averaging the decoding speed over a longer period, and in
that case you have lots of image buffers so the skip logic is less
likely to encounter the same image buffer again while the block is still
the same.